On 10/4/18 5:35 PM, Steve Wilson wrote:> On Thursday, October 4, 2018 at 12:38:20 PM UTC-4, Phil Hobbs wrote: >> On 10/4/18 11:39 AM, Steve Wilson wrote: >>> On Thursday, October 4, 2018 at 8:00:30 AM UTC-4, pcdh...@gmail.com >>> wrote: >>>>>> https://octopart.com/search?q=BFP650&oq=BFP650&start=0 >>>> >>>>> That would work, as a couple or few emitter followers to fan out >>>>> my incoming signal. It has decent beta at, say, 50 mA or even >>>>> 100 mA (driving something like 5 volts into 50 ohms). >>>> >>>>> The Vbe offset would be more predictable than the turn-on of a >>>>> phemt. >>>> >>>> Their low drain impedances make pHEMTs very disappointing as >>>> followers, unless you bootstrap the drain. The late lamented >>>> ATF38148 could have a gain of 0.5 as a follower. >>>> >>>>> I'll be timing edges to picoseconds so I don't want >>>>> unpredictable >>>> offsets. >>>> >>>> You'll need a base resistor to keep it from oscillating. Don't let >>>> anybody hang a cable on it. >>> >>> I hope that anyone playing with 42GHz GBW would automatically add a >>> base resistor to kill parasitic oscillations. Even a lowly 2N3904 can >>> go into parasitics with a long base lead. >>> >>> Often the real problem with parasitics is they can occur at a >>> frequency above your fastest scope, >> >> Well, I have a couple of 40-50 GHz ones, so that probably won't be an >> issue. ;) >> >>> so you can't see them. Sometimes even touching part of a circuit can >>> kill weak oscillations, making you believe the circuit is safe.> >>> Don't believe it. Add a base resistor to any circuit connected to the >>> outside world, or any place where fast transistors are connected >>> together. >>> >>> Often, a simple hand wave above a circuit can change the operating >>> conditions, >> >> I've posted a few times about my "manual wavemeter". From July 2015, >> concerning a pHEMT bootstrapped with a SiGe:C BJT: >> >> "The first proto oscillated at around 12 GHz. I didn't have a microwave >> spectrum analyzer at the time, so I measured the frequency by an >> interesting manual wavemeter method: if I held my hand over the board at >> different heights, the oscillation amplitude varied periodically, with a >> period of a bit less than half an inch." >> >> "The actual board was fine, after a bit of Dremelling and adding a bunch >> of parallelled bypass caps. With 40-50 GHz transistors and lots of gain >> in a small space, half a nanohenry here and there can ruin your whole day." >> >>> or a circuit may only go into oscillation at a certain point on the >>> input waveform. Both issues are hit or miss. >> >> AKA a snivet. >> >>> Add a base resistor. >>> >>> The value may range from 100 ohms for low frequency circuits, to >>> perhaps 5 or 10 ohms for fast circuits. You can easily change the >>> value as needed when you gain more experience with the circuit. But >>> it is extremely difficult to add a base resistor to a smd transistor >>> when you have forgotten to include it in the layout. Add a base >>> resistor. >> >> That blanket rule doesn't cover all the bases, so to speak. ;) Base >> resistors trash the speed and noise performance, so a tight layout and >> close attention to minimum-inductance bypassing help a lot. Then I use >> a zero-ohm jumper as a placeholder for a possibly-needed resistor or bead.> A zero ohm short may not be the best idea. You have to pay > > 1. to install the part. > > 2. to remove it when you find a problem. This can damage the pads. > > 3. to install a base resistor. This can also damage the pads.You only have to do it on the first article. Then you change the BOM and birds chirp.> > When you multiply the costs by the number of transistor involved, it can mean a substantial amount of money and time lost. > > If you install the resistor in the first place, you can eliminate these costs.And wind up with crap performance needlessly.> > After you gain some experience, you can pretty much tell how much resistance to use for each type of transistor and the kind of circuit used. I recommend using the highest value that won't degrade the performance of the circuit.Right, which is my zero-ohm jumper. ;)> > Also remember you may have a slow transistor in this production run. The next batch may be on the hot side. > > Here's a problem you may not see often. This was in an audio amp. The top trace in blue is the input signal. The bottom trace in red is the parasitic oscillation. > > https://drive.google.com/open?id=1o0YZiXfmORF-yvekNMp3W7Ny6iS8nJQgSure. Transistors get faster (up to a point) at higher I_C.> > Just because you are working in the audio range doesn't prevent a circuit from oscillating at VHF.The slowest scope I use regularly is 1 GHz, 4 GS/s, so those don't get by me. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
really fast buffers
Started by ●October 3, 2018
Reply by ●October 4, 20182018-10-04
Reply by ●October 4, 20182018-10-04
On 05/10/18 04:50, John Larkin wrote:> On Thu, 4 Oct 2018 08:39:11 -0700 (PDT), Steve Wilson > <9fe142ac@gmail.com> wrote: > >> On Thursday, October 4, 2018 at 8:00:30 AM UTC-4, pcdh...@gmail.com wrote: >>>>> https://octopart.com/search?q=BFP650&oq=BFP650&start=0 >>> >>>> That would work, as a couple or few emitter followers to fan out my >>>> incoming signal. It has decent beta at, say, 50 mA or even 100 mA >>>> (driving something like 5 volts into 50 ohms). >>> >>>> The Vbe offset would be more predictable than the turn-on of a phemt. >>> >>> Their low drain impedances make pHEMTs very disappointing as followers, unless you bootstrap the drain. The late lamented ATF38148 could have a gain of 0.5 as a follower. >>> >>>> I'll be timing edges to picoseconds so I don't want unpredictable >>> offsets. >>> >>> You'll need a base resistor to keep it from oscillating. Don't let anybody hang a cable on it. >> >> I hope that anyone playing with 42GHz GBW would automatically add a base resistor to kill parasitic oscillations. Even a lowly 2N3904 can go into parasitics with a long base lead. > > Or as an emitter follower with the base well bypassed to ground. The > wire bonds inside are nice high-Q inductors.How does an emitter follower do anything useful if the base is tied to ground?
Reply by ●October 4, 20182018-10-04
On Fri, 5 Oct 2018 09:26:17 +1000, Clifford Heath <no.spam@please.net> wrote:>On 05/10/18 04:50, John Larkin wrote: >> On Thu, 4 Oct 2018 08:39:11 -0700 (PDT), Steve Wilson >> <9fe142ac@gmail.com> wrote: >> >>> On Thursday, October 4, 2018 at 8:00:30 AM UTC-4, pcdh...@gmail.com wrote: >>>>>> https://octopart.com/search?q=BFP650&oq=BFP650&start=0 >>>> >>>>> That would work, as a couple or few emitter followers to fan out my >>>>> incoming signal. It has decent beta at, say, 50 mA or even 100 mA >>>>> (driving something like 5 volts into 50 ohms). >>>> >>>>> The Vbe offset would be more predictable than the turn-on of a phemt. >>>> >>>> Their low drain impedances make pHEMTs very disappointing as followers, unless you bootstrap the drain. The late lamented ATF38148 could have a gain of 0.5 as a follower. >>>> >>>>> I'll be timing edges to picoseconds so I don't want unpredictable >>>> offsets. >>>> >>>> You'll need a base resistor to keep it from oscillating. Don't let anybody hang a cable on it. >>> >>> I hope that anyone playing with 42GHz GBW would automatically add a base resistor to kill parasitic oscillations. Even a lowly 2N3904 can go into parasitics with a long base lead. >> >> Or as an emitter follower with the base well bypassed to ground. The >> wire bonds inside are nice high-Q inductors. > >How does an emitter follower do anything useful if the base is tied to >ground?I said "bypassed to ground." But an emitter follower can certainly work, and oscillate, if its base is really grounded. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
Reply by ●October 4, 20182018-10-04
On Thu, 4 Oct 2018 23:05:20 +0100, David Nadlinger <david@klickverbot.at> wrote:>On 03.10.18 11:03 PM, John Larkin wrote: >> Just resistors could work, but that loses amplitude, and the ends of >> the fanout traces would have to be always terminated to avoid >> reflections. Something unilateral would be easier. > >This doesn't exactly fit the "single-ended, easy termination" bill, but >I've recently had some fun (and success) with the ADCLK950: > >http://www.analog.com/media/en/technical-documentation/data-sheet/ADCLK950.pdf > >2 inputs, 10 outputs, propagation delay ~210 ps, 75 fs rms jitter.Nice, but it's all low-voltage differential. I need 5 volt swings in and out.> >How stable do you need the propagation delay to be?My DUT will be spec'd to 1 ns accuracy (well, the customer wants 200 ps, but that's silly) and maybe 30 ps RMS jitter. We'll have to occasionally characterize all the component and traces and cable delays in the test set, and fudge them out of the tests. Gigantic pain. After that, we'd like things to be stable to maybe 100 ps. I guess I should put an RTD or an LM35 on the test board, so my interfaced DVM can measure board temperature. I sure hope we'd never need to actually use that. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
Reply by ●October 5, 20182018-10-05
John Larkin <jjlarkin@highland_snip_technology.com> wrote:> On Thu, 4 Oct 2018 08:39:11 -0700 (PDT), Steve Wilson > <9fe142ac@gmail.com> wrote: >>I hope that anyone playing with 42GHz GBW would automatically add a base >>resistor to kill parasitic oscillations. Even a lowly 2N3904 can go into >>parasitics with a long base lead.> Or as an emitter follower with the base well bypassed to ground. The > wire bonds inside are nice high-Q inductors.How do you make an emitter follower when the base is bypassed to ground? Where do you put the signal?
Reply by ●October 5, 20182018-10-05
John Larkin <jjlarkin@highland_snip_technology.com> wrote:> On Fri, 5 Oct 2018 09:26:17 +1000, Clifford Heath <no.spam@please.net> > wrote:>>How does an emitter follower do anything useful if the base is tied to >>ground?> I said "bypassed to ground." But an emitter follower can certainly > work, and oscillate, if its base is really grounded.How? There's no signal on the base. It can certainly oscillate. But this is one place you would add a series base resistor to kill any parasitic oscillations.
Reply by ●October 5, 20182018-10-05
Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:> On 10/4/18 5:35 PM, Steve Wilson wrote: >> A zero ohm short may not be the best idea. You have to pay>> 1. to install the part.>> 2. to remove it when you find a problem. This can damage the pads.>> 3. to install a base resistor. This can also damage the pads.> You only have to do it on the first article. Then you change the BOM > and birds chirp.Then you find you have a problem in the field. You have to rework all the units you have shipped. Why not add a suitable base suppression resistor to critical circuits in the first place.>> When you multiply the costs by the number of transistor involved, it >> can mean a substantial amount of money and time lost.>> If you install the resistor in the first place, you can eliminate these >> costs.> And wind up with crap performance needlessly.I specified a value that does not degrade performance. There is absolutely no value in having a wider bandwith than you really need.>> After you gain some experience, you can pretty much tell how much >> resistance to use for each type of transistor and the kind of circuit >> used. I recommend using the highest value that won't degrade the >> performance of the circuit.> Right, which is my zero-ohm jumper. ;)>> Also remember you may have a slow transistor in this production run. >> The next batch may be on the hot side.> Cheers > > Phil Hobbs > >
Reply by ●October 5, 20182018-10-05
Am 05.10.2018 um 02:23 schrieb John Larkin:> On Thu, 4 Oct 2018 23:05:20 +0100, David Nadlinger> Nice, but it's all low-voltage differential. I need 5 volt swings in > and out.A real pain. And the receiver inputs after that will probably switch at 1.6V.> >> >> How stable do you need the propagation delay to be? > > My DUT will be spec'd to 1 ns accuracy (well, the customer wants 200 > ps, but that's silly) and maybe 30 ps RMS jitter. We'll have to > occasionally characterize all the component and traces and cable > delays in the test set, and fudge them out of the tests. Gigantic > pain. After that, we'd like things to be stable to maybe 100 ps.A customer of mine measured their cables in the temp chamber. That was real "fun". It is well-known that PTFE has a crystallization change at 23°C, where it hurts most, esp. when most quality cables are PTFE. If you want stable delays, it's probably best to try to be as fast as possible and then limit the BW with film R and NP0 C to sane values. Excess BW brings excess noise, brings excess jitter. Distributing a sine wave + a qualifier for the zero crossings takes cable dispersion out of the equation. One gains precisely located points in time but loses arbitrary edge position. regards, Gerhard
Reply by ●October 5, 20182018-10-05
On 10/4/18 11:21 PM, Steve Wilson wrote:> Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote: > >> On 10/4/18 5:35 PM, Steve Wilson wrote: >>> A zero ohm short may not be the best idea. You have to pay > >>> 1. to install the part. > >>> 2. to remove it when you find a problem. This can damage the pads. > >>> 3. to install a base resistor. This can also damage the pads. > >> You only have to do it on the first article. Then you change the BOM >> and birds chirp. > > Then you find you have a problem in the field. You have to rework all the > units you have shipped. Why not add a suitable base suppression resistor to > critical circuits in the first place. > >>> When you multiply the costs by the number of transistor involved, it >>> can mean a substantial amount of money and time lost. > >>> If you install the resistor in the first place, you can eliminate these >>> costs. > >> And wind up with crap performance needlessly. > > I specified a value that does not degrade performance.In your circuits, maybe. I often care very much about fractional nanovolts of noise.> > There is absolutely no value in having a wider bandwith than you really > need.In an audio amp, very true. In instruments, not so much. One of my main support headaches is people using the log output of the laser noise canceller, and adjusting the feedback bandwidth to set their measurement bandwidth. Causes all sorts of mayhem when the beam power ratios vary by order unity at frequencies near the feedback bandwidth. Running it fast fixes that, and it's easy to put an RC on the output if you like. It's usually best to have lots of bandwidth and then put in a good quality filter near the end. For instance, several of my gizmos use THS3091 CFAs as the output stage, configured as a two-pole Sallen-Key lowpass, and with a 50-ohm resistor in series with the output. The amp is 100 times faster than some of those circuits need, but it's super beefy (and hence student-resistant), and the extra speed keeps the S-K rolloff well controlled.> >>> After you gain some experience, you can pretty much tell how much >>> resistance to use for each type of transistor and the kind of circuit >>> used. I recommend using the highest value that won't degrade the >>> performance of the circuit. > >> Right, which is my zero-ohm jumper. ;) > >>> Also remember you may have a slow transistor in this production run. >>> The next batch may be on the hot side.Or it might have lower beta, in which case your procedure has the same problem. I generally go for about a factor of 1.5x to 2x more resistance than needed to cure the oscillation, and have never had an oscillation problem in the field. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
Reply by ●October 5, 20182018-10-05
On 10/4/18 11:09 PM, Steve Wilson wrote:> John Larkin <jjlarkin@highland_snip_technology.com> wrote: > >> On Thu, 4 Oct 2018 08:39:11 -0700 (PDT), Steve Wilson >> <9fe142ac@gmail.com> wrote: >>> I hope that anyone playing with 42GHz GBW would automatically add a base >>> resistor to kill parasitic oscillations. Even a lowly 2N3904 can go into >>> parasitics with a long base lead. > >> Or as an emitter follower with the base well bypassed to ground. The >> wire bonds inside are nice high-Q inductors. > > How do you make an emitter follower when the base is bypassed to ground? > Where do you put the signal? >The classical example is the one-transistor !RESET generator, which is a slow RC with an emitter follower. Cascading PNP and NPN followers to compensate V_BE is also a good way of making an oscillator. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
